FinRiskAtlas - Evidence-Grounded Processing - Quantitative Reasoning: leaderboard
FinRiskAtlas fixed-evidence review operation: computing a requested quantity from case figures and the applicable definition or formula, scored by numerical correctness (FinRiskAtlas, arXiv 2608.25325v1, Appendix Table 9). Percent; higher is better. Reported for 33 archived model configurations under zero-shot direct-answer inference.
Metric: Numerical Accuracy (%). Source: arxiv.org. Saturation forecast: Not forecast. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Ling-2.6-1T | 86.21 | #307 |
| 2 | GLM-5 | 84.48 | #149 |
| 3 | Nemotron 3 Ultra | 81.9 | #245 |
| 4 | Qwen 3 235B A22B | 81.03 | #323 |
| 5 | Qwen3-Next-80B | 80.17 | #321 |
| 6 | Qwen 3 32B | 79.66 | #449 |
| 7 | Seed 2.0 Lite | 79.31 | #161 |
| 8 | Seed 2.0 Mini | 79.31 | #279 |
| 9 | GLM-5.2 | 78.45 | #80 |
| 10 | Qwen 3.7 Max | 78.45 | #85 |
| 11 | GPT-5.6 Sol | 77.59 | #22 |
| 12 | Kimi K3 | 77.59 | #35 |
| 13 | GPT-5.4 Mini | 76.72 | #228 |
| 14 | DeepSeek V3 (0324) | 76.72 | #349 |
| 15 | GPT-5.6 Terra | 76.72 | #40 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=finriskatlas-evidence-grounded-processing-quantitative-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-05.